A database administrator reviews an AI-generated PostgreSQL performance fix awaiting human approval.
Microsoft's Azure Updates feed now lists "Public Preview: Database Agent for Azure Database for PostgreSQL." The label is a little generous. The detailed documentation describes a gated, region-limited preview, not one open to every Azure customer. Here is what is actually on offer and what you need before you can try it.

A database administrator reviews an AI-generated PostgreSQL performance fix awaiting human approval. Public preview, but with a velvet rope​

Azure Updates defines "In preview" as available to all Azure customers for non-production use and testing. The feature-specific documentation is narrower. Database Agents for Azure Database for PostgreSQL flexible server is currently available as a limited preview in Visual Studio Code. To request access, complete the Database Agents limited preview participation form.

A Microsoft Tech Community post on Azure HorizonDB puts it the same way. Database Agent for PostgreSQL is in Limited Public Preview, with access by request. Treat it as "public" in the announcement sense and "approval-gated" in the practical sense.

What the agent does​

The preview proactively monitors database health and investigates issues by correlating metrics, configuration, and workload signals. The scenarios Microsoft lists are connection exhaustion, dead tuple bloat, high CPU, idle in transaction, lock contention, query plan regression, and sort spill and memory pressure. Microsoft's FAQ describes these as the issues investigated on supported flexible servers.

Microsoft's documentation says the agent is built into the PostgreSQL extension for VS Code. From there you can configure an agent, review AI-assisted performance investigations, inspect live server state, and carry out manual responses without leaving the editor.

The Ignite 2026 Book of News adds the broader pitch. The agents use Microsoft-authored skills for common performance and health scenarios. They explain why a detected issue matters and recommend fixes or carry out approved actions. Microsoft says they work within existing identity permissions, approval boundaries and auditable actions. In its words, they support human judgment rather than replace it.

Prerequisites​

Check each of these before requesting access:

  • Approval: You must request access through the participation form and receive approval before enabling the feature.
  • Server: A running Azure Database for PostgreSQL flexible server on PostgreSQL 16 or later.
  • Region: South Central US, West US 3 or Sweden Central. During the limited preview, PostgreSQL support is available for flexible servers in those three regions only.
  • Compute tiers: Burstable, General Purpose and Memory Optimized tiers are included.
  • Editor: Visual Studio Code 1.102.0 or later is recommended for Model Context Protocol agent tools.
  • Extension: The PostgreSQL extension for VS Code, version 1.32.0 or later.
  • Permissions: A Microsoft Entra identity with access to the server resource, including Azure RBAC write permission on it when configuring the agent. You also need PostgreSQL credentials and network access to connect and run queries.
  • Optional: GitHub Copilot in VS Code, for the "Ask Copilot about this section" option and Agent mode tools.

On cost, Microsoft's FAQ says you don't pay for PostgreSQL Database Agents investigations during the limited preview. That could change when the preview ends.

Setting it up in VS Code​

  1. Select the PostgreSQL icon in the Activity Bar to open Connections.
  2. Under Servers, add or connect to your flexible server and sign in to Azure if prompted.
  3. Right-click the server. Choose Database Agent to configure it, or Dashboard to review active investigations and issues.
  4. If the dashboard shows Azure Server Detected or Incomplete Azure Metadata, select Fetch Metadata. Do the same if the agent won't save or activate. Then retry configuring Database Agent.

Reviewing a finding​

When the agent has detected issues, an Investigation banner shows the issue count. Expanding an issue shows its severity, status (for example DETECTED or DIAGNOSED), detection timestamp and AI-generated analysis. A recommended action can include:

  • Generated SQL, with Open in editor and Copy controls.
  • A "Why this is safe" explanation.
  • Rollback SQL.
  • An Expected outcome section with a rollback plan.

The issue buttons are Review action, Mark manually applied and Dismiss issue. The sparkle icon opens GitHub Copilot Chat with the issue details preloaded. You can then ask follow-up questions or inspect live metrics, sessions and execution plans.

The human stays in the loop​

Microsoft is blunt that evidence reflects the server at diagnosis time and that AI findings can be incomplete or incorrect. The documented workflow is:

  1. Confirm the condition is still live. Before terminating a backend, for example, query pg_stat_activity and pg_locks.
  2. For SQL changes, inspect the plan in the Query Plan Visualizer and review lock and resource impact. Microsoft gives CREATE INDEX CONCURRENTLY as a way to avoid blocking writes. Keep the rollback SQL ready, then run the approved statement manually through your authorized workflow.
  3. For parameter recommendations such as work_mem or autovacuum, right-click the server and go to Server Management > Server Parameters. If a static change needs a restart, use Server Management > Restart server in an approved maintenance window.
  4. Recheck live metrics and plans, then use Mark manually applied or Dismiss issue.

This is a suggestion engine with an audit trail, not an autonomous DBA. Anyone expecting it to quietly fix production at 3 a.m. will be disappointed. Change-control teams may find that reassuring.

What it isn't, yet​

  • Not in Fabric Database Hub for PostgreSQL. PostgreSQL Database Agents isn't yet available in Fabric Database Hub (coming soon). The Book of News says Database Hub itself is generally available, and that SQL and PostgreSQL agents there are in preview.
  • Not for Azure SQL Database in VS Code. Microsoft lists that support as coming soon. Azure SQL Database agents are available in preview in Database Hub.
  • Not for other PostgreSQL versions or hosts. Support covers flexible server on PostgreSQL 16 and later only.

Why admins should care​

Microsoft's own wording here is cautious: operators are meant to review and approve every change. For teams that live in VS Code, bringing diagnosis, query plans and server-parameter controls into one place could cut some context-switching.

Practical advice follows from the documentation. Start with a non-production server in one of the three regions. Confirm your RBAC write permission early, since that is a documented cause of agents failing to save. Treat the generated SQL as a draft, not a fix.

The caveat for planners is that this is preview software behind a request form. Don't build runbooks around it until access broadens and Microsoft publishes pricing and general availability terms.

 

References

  1. (In preview) Public Preview: Database Agent for Azure Database for PostgreSQL Azure Updates 2026-11-19T17:25:28Z
  2. database-agents-vs-code.md github.com
  3. Enable and activate Database Agents (Preview) - Microsoft Fabric learn.microsoft.com